Awesome World Model Hub Papers · Datasets · Projects
← Back to papers

Reward-free World Models for Online Imitation Learning

arXiv 24.10 2024 45.3 method

TLDR

Proposes reward-free world models for online imitation learning, using latent dynamics and inverse soft-Q learning to achieve expert-level performance on high-dimensional tasks.

Reasoning

The paper presents a novel integration of world models with imitation learning, addressing instability in high-dimensional tasks. Strengths include a clear methodology and diverse benchmarks (DMControl, MyoSuite, ManiSkill2). Weaknesses are the lack of explicit limitations or comparison details in the abstract.

Read-first score

Read-first score 45.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 32.

Recency 8%
75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Methodology quality 25%
50

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=benchmark

Topical relevance 42%
45.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%
30

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Candidate

Rank sensitivity

Stability: volatile; rank range: 313.

Keyword Scores

world model
10
world dynamics prediction
8
model-based reinforcement learning world model
6
generative world model
4
world simulator
2
interactive world model
1
video world model
1

Deep Analysis

Innovations

  • Reward-free world models for online imitation learning, learning environmental dynamics in latent spaces without reconstruction
  • Inverse soft-Q learning objective reformulated in Q-policy space to mitigate instability in reward-policy space optimization
  • Use of learned latent dynamics model and planning for control to achieve stable expert-level performance

Methodology

The method learns environmental dynamics entirely in latent spaces without reconstruction, adopting the inverse soft-Q learning objective to reformulate optimization in the Q-policy space. It employs a learned latent dynamics model and planning for control, enabling efficient and accurate modeling of high-dimensional inputs and complex dynamics.

Key Results

The approach consistently achieves stable, expert-level performance on diverse benchmarks including DMControl, MyoSuite, and ManiSkill2, demonstrating superior empirical performance compared to existing online imitation learning methods.

Tags